A Multilayered Preprocessing Approach for Recognition and Classification of Malicious Social Network Messages

Round 1

Reviewer 1 Report

congratulations on your work

please clarify your working method and add some findings or results to the conclusion.

for related work, I invite you to discover more work on user behavior on OSNs by different authors (idrais, sabour , ....etc) because it will be interesting to position the work well.

Author Response

In response to the feedback given by the reviewer, alterations have been made to the manuscript to offer increased clarity on certain aspects. 

Author Response File: Author Response.docx

Reviewer 2 Report

 

This paper is well presented and poses interesting work in the field due to it focusing on a specific language. However, what is not clear to me is why you don't simply translate the string and use one of the many already existing approaches in the english language. Moreover, I would want to see some kind of benchmarking in terms of comparing your approach to a different one using the translated (into english) string. I do understand, and enjoy you pointing out, the differences in the language you chose however it is not clear to me how you resolve contextual issues in which a bad keyword in a specific context means something positive. Finally, I am slightly (my personal opinion) apposed to papers that state, as you in your introduction, that the problem you are solving (harmful messages in social media) can be solved with a technical solution. We can help hybrid/human content quality and safety agents, but to solve it in the near future is highly unlikely. 

I would like to see more motivation and reasoning on why we should not simply translate the language into a highly researched language such as english. I assume that only needs a few lines, after which I am happy to accept the paper. 

As a side note, you do mention it in this line: This indicates that a multilingual generalization approach is probably to perform REF worse than an intralingual approach. 

However, I don't see why this indicates anything and I can't possibly agree to this either, unless there is a benchmarking performed.  

 

 

There are a good few typos and simple english mistakes for example - however they are very borderline and might even pass. 

This indicates that a multilingual generalization approach is probably to perform worse than an intralingual approach 

 

should be 

 

 

This indicates that a multilingual generalization approach probably performs worse than an intralingual approach 

but again, its very marginal.

Author Response

In response to the feedback given by the reviewer, alterations have been made to the manuscript to offer increased clarity on certain aspects. 

Author Response File: Author Response.docx

Reviewer 3 Report

Thanks for solving the problem of a Multilayered Preprocessing Approach for Recognition and Classification of Malicious Social Network Messages, but we have some remarks and questions that may help to highlight the work done:

1. Can you provide more details about the dataset used to train the model? How many data in the dataset did you train on?

 

2. It was mentioned that the data used to train the model is unbalanced. Did you learn by balancing the data? If so, can you explain the process? What techniques did you use to balance it out? If the data is not balanced, why? In Figure 2, why does adding up the 5 things add up to over 100%?

 

3. Malicious messages were classified using NB, KNN, and SVM methods. Why did you not use the deep learning method mentioned in related research?

 

4. If you look at how ‘accuracy>80%’ is handled in Figure 3, the picture and explanation do not match.

 

5. I need some explanation about Nr. Why does KNN have a particularly low accuracy rate in Aggressive?

Author Response

In response to the feedback given by the reviewer, alterations have been made to the manuscript to offer increased clarity on certain aspects. 

Author Response File: Author Response.docx